robo_set

  • 설명 :

38개 작업, 7500개 궤적에 걸쳐 12가지 중요한 조작 기술을 보여주는 단일 로봇 팔의 실제 데이터세트입니다.

나뉘다
'train' 18,250
  • 기능 구조 :
FeaturesDict({
    'episode_metadata': FeaturesDict({
        'file_path': string,
        'trial_id': string,
    }),
    'steps': Dataset({
        'action': Tensor(shape=(8,), dtype=float32),
        'discount': Scalar(shape=(), dtype=float32),
        'is_first': bool,
        'is_last': bool,
        'is_terminal': bool,
        'language_instruction': string,
        'observation': FeaturesDict({
            'image_left': Image(shape=(240, 424, 3), dtype=uint8),
            'image_right': Image(shape=(240, 424, 3), dtype=uint8),
            'image_top': Image(shape=(240, 424, 3), dtype=uint8),
            'image_wrist': Image(shape=(240, 424, 3), dtype=uint8),
            'state': Tensor(shape=(8,), dtype=float32),
            'state_velocity': Tensor(shape=(8,), dtype=float32),
        }),
        'reward': Scalar(shape=(), dtype=float32),
    }),
})
  • 기능 문서 :
특징 수업 모양 Dtype 설명
특징Dict
에피소드_메타데이터 특징Dict
에피소드_메타데이터/파일_경로 텐서
Episode_metadata/trial_id 텐서
단계 데이터세트
단계/작업 텐서 (8,) float32
걸음수/할인 스칼라 float32
걸음수/is_first 텐서 부울
걸음수/is_last 텐서 부울
단계/is_terminal 텐서 부울
단계/언어_지시 텐서
단계/관찰 특징Dict
단계/관찰/image_left 영상 (240, 424, 3) uint8
단계/관찰/image_right 영상 (240, 424, 3) uint8
단계/관찰/image_top 영상 (240, 424, 3) uint8
걸음 수/관찰/이미지_손목 영상 (240, 424, 3) uint8
단계/관찰/상태 텐서 (8,) float32
단계/관찰/state_velocity 텐서 (8,) float32
걸음수/보상 스칼라 float32
@misc{bharadhwaj2023roboagent, title={RoboAgent: Generalization and Efficiency in Robot Manipulation via Semantic Augmentations and Action Chunking}, author={Homanga Bharadhwaj and Jay Vakil and Mohit Sharma and Abhinav Gupta and Shubham Tulsiani and Vikash Kumar},  year={2023}, eprint={2309.01918}, archivePrefix={arXiv}, primaryClass={cs.RO} }